How Agentic AI Improves Security Operations

This is a live example of how I am using AI to help expedite cyber risk assessments at an enterprise level using an enterprise Claude environment. Let’s be honest. Regardless of how competent we are as cyber professionals, AI can often analyse faster, process larger volumes of information, and maintain consistency across assessments, provided it is given the right context, governance artefacts, and structured prompts. One of the persistent challenges in enterprise environments is the volume of cyber risk assessment requests. New integrations are introduced, new products are adopted, business units request policy exemptions, and security teams are expected to provide timely risk decisions. Using agentic AI, we can accelerate this process by enabling the model to retrieve the relevant governance artefacts, apply the organisation’s risk framework and matrix, map appropriate controls, draft the assessment, and prepare a structured report for consultant review. The result is not automation for the sake of automation, but a practical way to reduce manual effort while maintaining governance and oversight. Human expertise remains critical. The consultant still reviews the draft, validates the reasoning, and approves the final assessment. What changes is the speed and scalability of the process. The next evolution is even more interesting. Instead of building isolated agents for individual tasks, we should focus on building reusable skills that allow a single AI capability to orchestrate an entire governed workflow. As Anthropic has suggested, the future is not about building more agents, but about building the right skills that can be composed into reliable systems. Cyber GRC is entering a phase where AI will not just assist professionals, but help organisations execute governance processes faster, more consistently, and at enterprise scale. Thoughts? #CyberSecurity #CyberRisk #GRC #AI #AgenticAI #Claude #CyberGovernance #RiskManagement #EnterpriseAI #AIinCyber #DigitalTransformation

Are the magic words of AI-SOC agentic and autonomous? Triage, investigation, response, threat research, exposure management, malware analysis, and detection engineering are no longer islands. They behave like a connected reasoning graph, where each node feeds context to the next and pushes insights back into the system. The vision for an Agentic SOC makes this shift explicit. Instead of AI assistance, some of the model introduces autonomous multi agent systems that can analyze intent, break tasks into tasks, reason over evidence, and execute actions within guardrails. A collaborative system where humans lead and AI agents dynamically operate across the entire SOC surface. The real breakthrough is independence. Agents can identify when data is missing, request enrichment, correlate telemetry across domains, surface new hypotheses, and push improvements back into detection engineering. SOC work stops being a sequence of bottlenecks and becomes a feedback loop that continuously strengthens itself. Every alert, every artifact, and every hunt becomes learning material for the system. Data management becomes the backbone. Detection engineering becomes the learning engine. Triage becomes a reasoning hub, and incident response becomes the actuator for decisions born from a network of specialized agents capable of real analytical depth. I have dozens of them, but here are some principles for building an Agentic AI-SOC: Treat your telemetry as a trust contract. Agents cannot reason if the data is inconsistent, incomplete, or ungoverned. Operate detection engineering as a continuous reinforcement pipeline. Every investigation must feed back into what the SOC learns next. Give agents controlled autonomy. Let them correlate, enrich, hypothesize, and propose actions while humans own intent, boundaries, and oversight. One model to rule them all... not precisely, and you should have various models to behave at different tiers. The teams that adopt this model will operate at a level of efficiency and insight that traditional SOCs cannot achieve. The shift has already started. The question is who adapts and who gets left behind. #security #cybersecurity

🚨 Agentic Workflow for Insider Threat Monitoring 🧠🛡️ As enterprise data grows in complexity, insider threats are no longer just anomalies—they're sophisticated patterns that demand intelligent, context-aware monitoring. This cutting-edge Agentic AI architecture showcases how we can combine Machine Learning (ML), Large Language Models (LLMs), and rule-based automation to stay several steps ahead of potential security risks. 🔍 Key Highlights of the Workflow: 📥 Ingestion Layer: Seamlessly processes structured & unstructured security telemetry using Kafka, Amazon MSK, and Kinesis. 🧹 Preprocessing & Identity Mapping: Data Cleaner + PII Redactor (ML) ensures privacy by scrubbing sensitive information. Identity Graph Builder (ML) connects disparate user activities across systems to form a unified behavioral profile. 📊 Behavioral Analysis & Anomaly Detection: Baseline Behavior Modeler (ML) establishes “normal” behavior for every identity. Anomaly Detection Agent (ML) flags deviations using ML guardrails for precision and accountability. 🤖 Agentic Intelligence (LLM + Rule Engine): Threat Synthesizer Agent (LLM) reasons over anomalies and combines contextual signals from vector databases like Pinecone, Weaviate, and Amazon OpenSearch. Soar Executor Agent triggers appropriate actions using pre-set rules. Feedback Interpreter & Learner (LLM) learns from analyst feedback and continuously improves threat detection. 🧠 LLM Infra: Powered by Amazon Bedrock, OpenAI, and Claude 3 Sonnet—providing the scale and intelligence needed for complex, real-time decision making. 📈 Transparency & Explainability Tools: Integration with SageMaker Clarify, EvidentlyAI, and Bedrock Guardrails ensures fairness, transparency, and compliance. 💬 Human-in-the-loop: Analysts can review and interact through tools like Slack, Jira, and a dedicated Analyst Interface for final verdicts or overrides. 🔐 This isn’t just automation—it's augmented security intelligence, capable of evolving with your threat landscape.

Agentic AI is reshaping the attack surface. These systems don't just answer questions; they reason, take action, and interact with data and tools. That autonomy brings risk, especially when prompt injection or task hijacking can quietly redirect an agent's decisions. For CISOs, the answer isn't to slow innovation. It's to design guardrails that match the stakes. Limit what agents can access, treat them as identities with least-privilege controls, and monitor their behavior the same way you would any other system making decisions on your behalf. AI agents can accelerate security workflows when used with discipline and clarity. Strong oversight turns them into an asset; neglect turns them into an exposure. #Cybersecurity #CISO #AIThreats

🚀 Agentic AI and the need for Zero Trust Security Over the past couple of days I got questions about the security side of Agentic AI. When we talk about AI agents that can access business tools, sensitive databases, and internal APIs, security can’t just be an afterthought- it has to be the starting point. That’s where zero trust comes in. Zero trust is not just a tech buzzword. It’s a simple but powerful idea: don’t automatically trust anything or anyone - inside or outside your company’s systems. Always verify, every single time. So, what does zero trust actually look like? Here are a few features that define it - and why they matter so much for Agentic AI: 1️⃣ Never Trust, Always Verify: Every request - whether it’s an AI agent trying to fetch data, or a user logging in - must be checked and validated. Nothing is “trusted” just because it’s inside the network or from a familiar system. 2️⃣ Least Privilege: Only give access that’s absolutely needed. If an AI agent just needs to read sales numbers, it shouldn’t have access to edit or delete data. Permissions are tightly controlled, and kept as limited as possible. 3️⃣ Continuous Authentication: It’s not “log in once and you’re good.” Every action, every API call, every data request is checked. Tokens are short-lived, credentials are rotated, and the system is always asking, “Are you still allowed to do this?” 4️⃣ Micro-Segmentation: Even within your systems, different tools and data sources are separated into small “segments.” The AI agent has to prove it has the right to cross into each one—it’s never an all-access pass. 5️⃣ Audit and Monitoring: Everything the agent does - what it accesses, what tools it uses, what data it pulls - is logged. This isn’t just for compliance, but for spotting mistakes or suspicious behavior quickly. 6️⃣ No Hardcoded Secrets: Agents should never have passwords or API keys baked into their code. Use secure vaults or secret managers, and make sure everything is protected and easy to rotate. Why is all this so relevant for Agentic AI? Because these agents are smart and fast - they can access multiple tools in seconds and scale their actions without much human intervention. If you don’t put strong controls in place, a small mistake or security gap can lead to a big problem. So if you’re building, deploying, or even just experimenting with Agentic AI, start with zero trust. Treat every agent as you would an external visitor. Always ask: 👉 Should this agent have access right now? 👉 Is it doing only what it’s supposed to do? 👉 Can I see and control everything it touches? Sometime, I have be been challenged by a few if this will slow down innovation - my answer is a definite "No". In fact, it’s what lets you move faster, knowing your data and systems are protected at every step. I write about #artificialintelligence | #technology | #startups | #mentoring | #leadership | #financialindependence   PS: All views are personal Vignesh Kumar

The Rise of Agentic AI in Cybersecurity Last year, I worked with a security team overwhelmed by a flood of alerts caused by a misconfigured firewall. Thousands of notifications came in overnight, and by the time analysts sorted through the data, the real threat had already taken hold. Today, that story could look very different—thanks to Agentic AI. Agentic AI doesn’t just assist security teams; it acts autonomously. It identifies and groups alerts, adds context, hunts threats using frameworks like MITRE ATT&CK, and can even take action—such as isolating devices or updating firewall rules—without waiting for human input. The benefits are clear: Faster, more effective threat detection Reduced alert fatigue and burnout Improved team morale and retention From triage and investigation to automated penetration testing, Agentic AI is changing how organizations secure their systems. But there are challenges to solve: transparency, data quality, false positives, and the need for strong human oversight. As this technology becomes more integrated, the question becomes not if we should adopt it—but how. Can we trust autonomous AI to take meaningful action in cybersecurity, or will human decision-making always be a necessary part of the equation? #CyberSecurity #AgenticAI #AIinSecurity #ThreatDetection #Automation #InfoSec #FutureOfWork

Banks New Operating Model Could Move from Humans to Al Agents at the Center 💡 Agentic AI is often best for complex tasks requiring proactive execution and autonomous decision-making while traditional AI may suit pattern recognition and predictive tasks. Banks should deploy each where it adds the most value. Banks should proactively embed compliance considerations directly within the AI agents’ operational logic, workflows, and oversight mechanisms. Establishing built-in compliance guardrails, automated risk assessments, and continuous monitoring should help agents to operate more securely within the compliance frameworks. However, this can only happen with close collaboration between the compliance teams and the AI development groups in both the design and deployment phases, which can result in more transparent, explainable, and accountable agentic AI 🤖 Supporting agentic AI will likely require robust and scalable infrastructure. Cloud-based solutions can offer the flexibility and computational power often necessary for these systems to operate efficiently. Additionally, as AI agents proliferate, orchestrating a multi-agent system in a seamless manner become crucial. Agentic AI should use high-quality, accessible data for a better chance at success. Banks should strengthen data governance, emphasizing consistent data standards, quality control, and comprehensive metadata management. Establishing robust data pipelines and centralizing assets into data lakes or fabrics should enable more seamless agent access and rapid analytics. Equally critical is robust data lineage that can maintain data integrity and enhance the explainability and transparency of agents’ decisions 🔍 As the number of AI agents expand, it would be prudent to maintain a comprehensive and updated agent registry with information on owner(s), scope, data sets, and risk exposure limits of every agent. The shift to agentic AI will likely require a significant cultural change, more so than traditional AI or standard LLMs. Rather than positioning humans at the center of task execution, banks should adapt to a model where AI agents increasingly take on a central role, with humans guiding, overseeing, and intervening as needed. In the near term, banks should be selective and focus on areas that can provide more immediate value. Agentic AI systems may take some time to yield robust gains, but these benefits may differ by the maturity of the institutions on data, cloud, and AI infrastructure and capabilities. In addition, the benefits of agentic AI such as speed and efficiencies should be balanced with risk considerations. Banks should navigate risks such as data privacy, ethical dilemmas, and regulatory compliance. Proactive experimentation and balanced risk-reward trade-offs are likely to be crucial for success 🔥 Source: Deloitte - https://shorturl.at/EQ7fo #Innovation #Fintech #Banking #FinancialServices #AI #Data #Cloud #LLMs #GenAI #Agents #Compliance #Strategy

Reading the new Agentic AI Identity and Access Management report from the Cloud Security Alliance made me pause. It highlights something we often overlook. Thats the the fact that existing identity systems were never designed for autonomous agents. These agents do not just log in like humans or service accounts. They make decisions, interact across multiple systems, and act in ways that traditional IAM simply cannot handle. Key highlights from the report • Traditional protocols like OAuth, OIDC, and SAML fall short in multi-agent environments because they assume static identities and predictable workflows • AI agents require fine-grained, context-aware permissions that change in real time • Agent IDs based on Decentralized Identifiers and Verifiable Credentials allow provenance, accountability, and secure discovery • The proposed framework blends zero trust principles, decentralized identity, dynamic policy enforcement, authenticated delegation, and continuous monitoring • Concepts like ephemeral IDs, just-in-time credentials, and zero-knowledge proofs address the privacy and speed demands of autonomous systems Who should take note • Security leaders preparing for agent-driven enterprise systems • Engineers and architects designing secure frameworks for agent-to-agent communication • Product teams deploying agents into sensitive workflows • Governance leaders shaping accountability and compliance policies Why this matters Our identity models were built around human users and predictable software. Agentic AI changes that equation. Without new approaches, we risk security blind spots, accountability gaps, and over-privileged systems that cannot be traced or revoked in time. The path forward Enterprises need to start treating AI agents as first-class identities. That means verifiable credentials, continuous monitoring, and dynamic delegation as the baseline. This is not about adding more controls. It is about reshaping IAM so that trust, security, and accountability are preserved in the age of autonomous systems.

While AI SOC dominates headlines, security engineering teams are quietly grappling with a 40% annual surge in security data volume. That’s why I’ve long stressed the growing importance of the Data ETL/pipeline market—one of the most critical, yet overlooked, aspects of the SOC. Today, rather than just using AI SOC for incident response triage, we’re seeing a new trend: AI is transforming how SOC engineers process, manage, and extract value from their data. A recent announcement I saw from Observo AI highlights this transformative trend. For context, for non-SOC folks, traditional security data pipelines require specialized engineering expertise, deep knowledge of query languages on Splunk, and time-consuming manual effort. As a result, security teams often face delays in investigation and response, despite having access to large amounts of data. Observo AI just launched (Orion AI). This is one of the first case studies where AI is leveraged to address data pipeline issues. Along with its agentic AI-based platform, Orion AI functions as an AI-powered data engineer, allowing security and DevOps teams to ingest, route and manage data pipelines from multiple sources, optimize workflows, standardize, enrich, correlate, normalize and query cloud-stored data—all through natural language. Some case studies of how we're seeing AI being leveraged in security engineering and what I've seen with Orion AI: 1) Data Pipeline Automation - AI can enable teams to define end-to-end pipelines from multiple sources to multiple destinations through an LLM-based conversational interface. 2) AI-Powered Querying & Search - AI can allow security teams to search and interact with live and archival data using natural language, eliminating the need for complex and proprietary queries. 3) Pipeline Optimization & Cost Efficiency - Machine learning identifies inefficiencies in data processing and reduces storage costs in real-time, while maintaining observability. 4) Interactive Pipeline Management - Provides real-time control over security and observability data pipelines through Agentic AI. 5) Incident Response Acceleration - Streamlines access to security-relevant data, reducing investigation times by 40%+ Why do I think security leaders and engineers should care? IMO, security teams shouldn’t be blocked by data bottlenecks or a reliance on specialized engineers just to extract insights. AI is now able to shift the paradigm by making security and observability data more accessible, actionable, and cost-effective. The question now is: How should security teams integrate AI into their workflows to improve efficiency without compromising control? *** PS: I'll be sharing much more about how AI is being leveraged in the SOC (not for triage, but more so within the data engineering pipeline by the end of March. See the comments to subscribe if interested in this topic)

Most security frameworks were built for a world where software does exactly what you tell it to do, every time. Agentic AI breaks that assumption. Agents use LLMs to carry out actions on their own, at machine speed, with real-world consequences. And because they’re non-deterministic, the same request can produce different results each time. That’s a fundamentally different operating model, and it raises questions our industry needs to answer well. NIST’s Center for AI Standards and Innovation recently issued a Request for Information asking for industry input on how we should secure these systems. We submitted a response based on our experience building and operating agentic AI services at AWS, and we published a blog summarizing the four security principles at the core of it. A few points I’d emphasize for anyone thinking about how to secure agents at their own organization: 1. Secure foundations are more important than ever. Every traditional attack technique, including denial-of-service, man-in-the-middle, vulnerability and configuration exploitation, supply chain, log tampering, etc., remains relevant in agentic contexts. AI-specific controls must be additions to foundational security, not replacements for it. 2. Don’t rely on the agent to secure itself. Even if you tell an LLM to refuse certain requests, crafty prompt injection techniques can override those instructions. Security boundaries need to be enforced by infrastructure outside the agent that governs what it can access and do. And these controls must be deterministic. 3. Autonomy should be earned, never granted by default. Start by having humans make the final call on high-consequence operations. As you gather evidence that the agent performs reliably, expand its autonomy gradually. And be ready to pull it back when the data says you should. 4. Be thoughtful about human-in-the-loop oversight. If every action requires approval, reviewers get overwhelmed and start rubber-stamping. Focus human oversight on the decisions that genuinely carry high stakes. We’re all figuring this out in real time, and no single organization has all the answers. The more we share what we’re learning, the faster the whole industry moves forward. For more details on how to apply these principles, check out the links below. Full response to NIST: https://lnkd.in/enxE8R-V Blog post: https://lnkd.in/eRg3uc26